---

name: suggest_model description: 次のタスクの内容（コーディング、リサーチ、リファクタリングなど）に応じて、最適なLLMモデルを提案する

Suggest AI Development Tool & Model

Suggest the optimal combination of models and tools from the available development environments (Google Antigravity, OpenAI Codex, Gemini CLI, VS Code) based on the nature of the task.

1. Available Models and Environments (as of 2026)

Google Antigravity (Google One AI Pro plan)

• Gemini 3 Pro (High/Low): High precision, deep reasoning, and hyper-large context. Best for repository-wide analysis and complex designs.
• Gemini 3 Flash: Rapid response. Best for small-scale fixes and batch processing.
• Claude Sonnet 4.5 (with/without Thinking): General-purpose model with an excellent balance of reasoning and generation. The "Thinking" version is for complex logical construction.
• Claude Opus 4.5 (Thinking): Top-tier analytical power. Best for strict physical/mathematical verification and difficult bug fixes.
• GPT-OSS 120B (Medium): Large-scale model based on open source.

OpenAI Codex (Team plan / CLI / IDE Extension)

• GPT-5.2 / GPT-5.2-Codex: Specializing in the latest engineering. For large-scale refactoring and tool integration.
• GPT-5.1-Codex-Max: Long-duration autonomous execution and sustained large-scale tasks. Includes context compression for end-to-end execution.
• GPT-5.1-Codex-Mini: For small-scale, rapid autonomous fixes.

Gemini CLI (Free Trial)

• Gemini 2.5 Pro / Flash / Flash Lite: For lightweight tasks via CLI or quick investigations.

VS Code (GitHub Copilot / Ollama)

• GitHub Copilot: Real-time completion.
• Ollama (llama3.1:8b, qwen2.5:7b, gemma3:4b, etc.): For local execution, highly sensitive experiments, and offline work.

2. Recommended Usage

1. Large-scale structural changes / Physical-mathematical implementations: Claude Opus 4.5 (Thinking) or Gemini 3 Pro (High)
2. Long-duration, complex autonomous end-to-end execution: GPT-5.1-Codex-Max
3. High-speed coding / Documentation: Gemini 3 Flash or Claude Sonnet 4.5
4. Local confidential prototyping: Ollama (qwen2.5:7b / llama3.1:8b)

3. Procedure

1. Identify Task Category: Determine whether the primary purpose is analysis, implementation, testing, or documentation.
2. Consider Resource Status: Select the optimal environment considering the quota (High/Medium/Low).
3. Proposal: Present the name of the model and environment along with specific advantages.